Segmentation of cell nuclei in tissue by combining seeded watersheds with gradient information

Abstract

This paper deals with the segmentation of cell nuclei in tissue. We present a region-based segmentation method where seeds representing object- and background-pixels are created by morphological filtering. The seeds are then used as a starting-point for watershed segmentation of the gradient magnitude of the original image. Over-segmented objects are thereafter merged based on the gradient magnitude between the adjacent objects. The method was tested on a total of 726 cell nuclei in 7 images, and 95% correct segmentation was achieved. © Springer-Verlag 2003.